See
http://www.reddit.com/r/programming/comments/7bse3/david_roundy_steps_down_as_maintainer_of_darcs/c0686jp

I think a better option for trackdown (which never did get much attention) would
be to use the delta debugging approach, which is a generalization of the
bisection approach that works with sets of changes rather than a linear sequence
of versions.

For the interested, I think it was Jason that first told me about Delta
Debugging.  I didn't fully understand it, but it sure sounded cool:
http://en.wikipedia.org/wiki/Delta_Debugging

+1, this would be useful for me.

I think we need somebody to write up a description of how delta debugging would
work in a darcs trackdown setting.

I've split the Delta Debugging stuff into a new ticket, issue1638.

While I imagine that Delta Debugging would be really very cool, I don't want us
to get stuck on it when we could very easily implement bisection, which would
already be useful on its own.

I'm also bumping the priority and marking this easy because I think it's
localised to just the trackdown command.

Radoslav and me are working on this, patch is on the way...

it is done.  it's not pretty, and it's cheating with unsafeUnFL/RL, but it does
what it is supposed to do on our test cases.  if you reject it please be nice
about it.  i know unsafeUnFL/RL needs to be eliminated, but my brain is not
working any more, and i need to be told that we've done something useful today.  :-)

As we discussed at vienna sprint, it seems it would be the best re-write it
using builtin Haskell Tree data structure, what can make it more readable/correct.

hi radoslav, look at the patch: i did that already.  (-:

sorry i didn't notify you explicitly any more.  after we found the
error and we fixed it, i did some clean-up and committed everything we
did as a single patch.

but it still needs to be rewritten.  see FIXME paragraph in the latex
documentation in the same file.

now i'm back at work.  i'll get back to this later this week.  feel
free to steal the work away from me!  :) or ask me if you think i can
help you.

matthias



On Mon, Nov 16, 2009 at 10:02:22AM +0000, Radoslav Dorcik wrote:
> To: darcs-devel@darcs.net, dixiecko@gmail.com, dmitry.kurochkin@gmail.com,
> 	jjenning@fastmail.fm, kowey@darcs.net,
> 	mf-hcafe-15c311f0c@etc-network.de, naur@post11.tele.dk,
> 	trentbuck@gmail.com
> From: Radoslav Dorcik <bugs@darcs.net>
> Date: Mon, 16 Nov 2009 10:02:22 +0000
> Subject: [issue1208] wish: darcs trackdown --bisect
> 
> 
> Radoslav Dorcik <dixiecko@gmail.com> added the comment:
> 
> As we discussed at vienna sprint, it seems it would be the best re-write it
> using builtin Haskell Tree data structure, what can make it more readable/correct.
> 
> __________________________________
> Darcs bug tracker <bugs@darcs.net>
> <http://bugs.darcs.net/issue1208>
> __________________________________
> 
> 
>  ** ACCEPT: CRM114 PASS osb unique microgroom Matcher **
> CLASSIFY succeeds; success probability: 0.9471  pR: 1.2531
> Best match to file #0 (nonspam.css) prob: 0.9471  pR: 1.2531
> Total features in input file: 2048
> #0 (nonspam.css): features: 761191, hits: 2639451, prob: 9.47e-01, pR:   1.25
> #1 (spam.css): features: 1678863, hits: 2969065, prob: 5.29e-02, pR:  -1.25
>

Hi, here is updated patch without unsafeFL, now bisect functionality uses FL
instead of []. The True/False flag in generate method parameters have to remain
because FL/RL types itsef can not replace that information included in
True/False flag. 

Actually this thing was one in your TODO. I hope I'll find the time to create
also module tests (or automatic tests) for bisect.

Attachments

• issue1208-bisect-patch-with-unsafeFL-removed.dpatch

hi, http://bugs.darcs.net/patch106 obsoletes all other patches posted here
earlier, and is in my opinion ready for review:

 - i took radoslav's improved generate function, massaged it a bit more, 
   and now it's quite easy to read and compliant with the GADT
   conventions.

 - i added a unit test script and repositories.

--match, --last, etc. are still missing, but i think that should be a new
patch that can be reviewed independently.  (i'm currently looking at
how diff does it, and i think i can just copy a few lines from there.
any opinions on that approach?)

thanks / cheers,
matthias

(i changed the status to "waiting-for".  not sure if this means "needs review"?)

Resolved with this patch (but our auto-resolver got confused by the dash).

Congrats to Rado and Matthias.  I think darcs trackdown just became more
useful :-)

Mon Mar 29 23:10:33 BST 2010  dixiecko@gmail.com
  * resolve-issue1208: trackdown --bisect (PatchTree).